Ran UChicago Proteomics Core for 10+ Years
15+ years of Proteomics Experience
5+ tools for sorting/analyzing/plotting protein data (Prism, DAVID, STRING, Maxquant, etc.)
Hasn't graduated high school yet
Learned what proteomics was in the summer of 2023
Has access to OpenAI's GPT4
Overall rationale: There is strong evidence in the literature that cellular responses a critical time window over a few day after bone marrow damage from radiation or other genotoxic stress can determine delayed onset disease including bone marrow failure and cancers. We propose that key process may be the formation and persistence of senescent cells, leading to a chronic inflammatory response that drives long-term bone marrow damage, genomic instability and malignance. By acting early to promote DNA repair and suppress onset of senescence, it may be possible for the bone marrow to fully recover before the destructive process can get underway.
Search Data using
Maxquant at 1FDR (Against Mus Fasta, with Secreted against (Mus, Bovine, Feline Fastas).
NR/Ctrl for (IR, NIR, IR_Sec, NIR_Sec)
Median Normalize Data sets
Search Upregulated proteins above Log2 >= 0.26
Reactome for Super Pathway enrichment (Plot– log10FDR, #Hits (Proteins)) https://reactome.org/
gProfiler for (GOBP, GOMF, GOCC, and KEGG) https://biit.cs.ut.ee/gprofiler/gost
Plot –log10(Adj P-value or FDR)
For Bone Marrow IR Pathway, pull out pathway hits and plot XY (IRNR/IR x-axis, NIRNR/NIR y-axis)
Immune Response
Redox or Oxidative Stress
Stress Response (“Stress”)
Apoptotic Cell Death
DNA Damage
DNA Repair
Chromatin Organization
For interesting hits pull out pathway genes and use String-db to obtain network
NIRNR_NIR:
Specific pathways that showed the most hits with their respective adjusted p-values
NIRNR_NIR:
Specific pathways that showed the most hits with their respective adjusted p-values
Human Made:
Protein hits for each bone marrow test
Human Made:
Venn Diagram with the hits that were found within the various tests
AI Made:
String that links all of the genes involved with DNA repair
AI Made:
Heatmap of upregulation of genes involved within DNA repair found in treatments
We were able to have Chat GPT-4 create a formula for finding the "most significant" biological processes using both the hits and the adjusted p-value. Next, it cross referenced these significant processes with acute radiation syndrome and made a hypothesis for what could be most important to look at within the genes themselves. (Keep in mind that GPT-4 lies a lot and is limited to September 2021)
Can save a lot of time with creating graphics or isolating specific datapoints
Can access databases and link specific proteins to specific pathways (not 100% accurate/useful all the time)
Limits the use of going in between tools (Venny, String, Reactome, etc)
Can upload data files and will reliably sort them
Can falsify data very easily and give fake sources (usually if it gives a methodical response, its b.s.)
"in order to answer this prompt, one would do x or y to get z" : z is most likely made up
50 prompts every 3 hours with GPT-4
Can "forget" reference files from the conversation even within the same tab
Cannot access recent information
It was written to please the user
I suggest that the best way to treat current AI technology is to act as if it was a computer science major that never made it past Bio102
In the TET project, GPT-4 was mostly used as a planner/protocol creator for following experiments. However, I did have to constantly ask for sources and links to make sure that the information it gave me was not made up. It is also very effective as a code writer or graph maker, but those skills were not needed in the TET project.
While AI is an exciting new tool that is being made available. the fact that it can lie about information to please the user is quite alarming. The best way that I was able to use this new technology was as a code/graph builder, rather than a teacher or creative thinker.